#!/bin/bash
#SBATCH --account=fmri
#SBATCH --partition=g40x
#SBATCH --job-name=memoryrr
#SBATCH --nodes=1              
#SBATCH --ntasks-per-node=8     # should = number of gpus
#SBATCH --gres=gpu:8
#SBATCH --time=32:00:00          # total run time limit (HH:MM:SS)
#SBATCH -e slurms/%j.err
#SBATCH -o slurms/%j.out
#SBATCH --comment=fmri



export NUM_GPUS=8  # Set to equal gres=gpu:#!
export BATCH_SIZE=32
export GLOBAL_BATCH_SIZE=$((BATCH_SIZE * NUM_GPUS))

# Make sure another job doesnt use same port, here using random number
export MASTER_PORT=$((RANDOM % (19000 - 11000 + 1) + 11000)) 

export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
export COUNT_NODE=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | wc -l)

export WANDB_DIR="/fsx/proj-fmri/ckadirt/MindEyeV2/src/wandb" 
export WANDB_CACHE_DIR="/admin/home-ckadirt/.cache"
export WANDB_MODE="online"

echo MASTER_ADDR=${MASTER_ADDR}
echo MASTER_PORT=${MASTER_PORT}
echo WORLD_SIZE=${COUNT_NODE}

source /admin/home-ckadirt/.bashrc

###########

cd /fsx/proj-fmri/ckadirt/MindEyeV2/src/
accelerate launch --num_processes=$(($NUM_GPUS * $COUNT_NODE)) --num_machines=$COUNT_NODE --main_process_ip=$MASTER_ADDR --main_process_port=$MASTER_PORT Train_MLPMixer-img.py --data_path=/fsx/proj-fmri/shared/mindeyev2_dataset --model_name=testing-rr-1024-img-past-2 --subj=1 --batch_size=${GLOBAL_BATCH_SIZE} --max_lr=3e-4 --mixup_pct=.66 --num_epochs=120 --ckpt_interval=10 --no-use_image_aug --hidden_dim=1024 --seq_len=2

# --wandb_log